Accidents that result in immediate fatalities and injuries leading to disabilities occur frequentlyaround the globe. Dependents of claimants are able to seek compensation through the legal system,a process that can take up to 5 to 6 years. Our proposed model, an Adaptive Weighted ExtremeFusion Regression Model (XFRM-Justice), automates the prediction of compensation awarded bycourts. The state of the art Regressors and classical ML algorithms such as the Random Forest, theSVM and some of the other algorithms can deliver competitive performance but do not readilyextend to other patterns of judgment. To solve these shortcomings, we proposed a new hybridmachine-learning system, XFRM-Justice, a hybrid adaptive weighted extreme fusion regressor isan enhanced hybrid of XGBoost, CatBoost, LightGBM and ExtraTrees, and applied an HybridAdaptive Weighted Fusion (AWF) algorithm to merge the output of these regressors. A 5-foldcross-validation approach with internal validation is used in order to guarantee robustness. Thedataset utilized is MVOP, derived from the original MVOP judgments from the Indian Kanoonrepository and various districts e-courts of India, which consists of 16 features. We evaluated theperformance of the models using metrics such as MSE, MAE, RMSE, and R² and compared withindividual ensemble base, ablation models. AWF actively determines the best weights eachregressor should have based on its R2 value, Mean Absolute Error (MAE) and also its Root MeanSquare Error (RMSE) score, yielding a situation-dependent and very robust fusion. Using theMVOP dataset, XFRM-Justice shows high performance in terms of MSE, MAE, RMSE, and R² ,achieving an MSE of 755665302970.5956, MAE of 205492.1799, RMSE of 825021.7851, and anR² of 0.8868. These results indicate that the Proposed model is highly effective for developing areliable prediction system for legal judgments.
Prasanna Kumari Paturu (Wed,) studied this question.